Revisiting Read Wear: Analysis, Design, and Evaluation

Home , Scrollbar

Revisiting Read Wear: Analysis, Design, and Evaluation of a Footprints Scrollbar Jason Alexander 1, Andy Cockburn 1, Stephen Fitchett 1, Carl Gutwin 2, Saul Greenberg 3 1 Dept. of Computer Science 2 Dept. of Computer Science 3 Dept. of Computer Science University of Canterbury University of Saskatchewan University of Calgary Christchurch, New Zealand Saskatoon, Canada Alberta, Canada {jason, andy, saf75} [email protected] [email protected] @cosc.canterbury.ac.nz

ABSTRACT having attended to the thumb’s location during previous In this paper, we show that people frequently return to visits, remembering it, and reproducing it accurately. One previously-visited regions within their documents, and that or more of these activities can fail. In their comparison of scrollbars can be enhanced to ease this task. We analysed navigation in paper and electronic documents O’Hara and 120 days of activity logs from Microsoft Word and Adobe Sellen [25] observed that people use fingers to mark Reader. Our analysis shows that region revisitation is a document regions in paper documents for easy return, and common activity that can be supported with relatively short that equivalent methods are lacking when scrolling. They recency lists. This establishes an empirical foundation for conclude that scrolling is “irritatingly slow and distracting” the design of an enhanced scrollbar containing scrollbar and that people need “quicker, more effortless navigation” marks that help people return to previously visited [25], p.341. document regions. Two controlled experiments show that scrollbar marks decrease revisitation time, and that a large One way to improve support for revisitation is by number of marks can be used effectively. We then design augmenting the scrollbar region with scrollbar marks. Such an enhanced Footprints scrollbar that supports revisitation marks are not a new idea. Attribute-mapped scrollbars , with several features, including scrollbar marks and mark patented in 1990 [31], used coloured marks in the scrollbar thumbnails. Two further experiments show that the to draw attention to salient properties; scrollbars are well- Footprints scrollbar was frequently used and strongly suited to showing this information, as they provide an preferred over traditional scrollbars. overview of the entire document. Hill et al. [18] used a similar approach to denote the read wear that occurs with Author Keywords. use – the marks portray how often sections of the document Document revisitation, read wear, scrolling. have been read. However, their focus was on showing history of use, not on supporting revisitation. ACM Classification Keywords. Read-wear marks on the scrollbar offer a lightweight H.5. Information interfaces and presentation: User method for improving document revisitation compared to interfaces ( Interaction styles , screen design ). traditional scrolling: a mark shows a person where they have been, and provides a navigation cue to help them INTRODUCTION quickly return to that spot. However, the idea has not Scrollbars are a familiar widget common to graphical user caught on, and we are unaware of any system currently interfaces, and have become a standard idiom for view using it (although scrollbar marks are becoming common in navigation within documents. They are compact and IDEs for marking code errors or comments). conceptually simple, yet powerful, providing useful information about the viewport’s location and extent in a We believe the poor adoption of read-wear scrollbars is due larger information space. In particular, the scroll thumb’s to a lack of knowledge about how revisitation occurs in the spatial cue can help revisitation – returning to previously real world, how best to design a read-wear scrollbar, and visited document regions. For example, a user may know the potential benefits and harms of using it in realistic that “moving the thumb roughly four-fifths of the way systems. Consequently, we conducted several investigations down will bring me to the Results section”. to address these knowledge gaps. However, rapid and effective revisitation depends on people 1. How do people revisit document locations, and are current tools well used? We analysed logs from a 120- Cite as : Alexander, J., Cockburn, A., Fitchett, S., Gutwin, C., and day longitudinal study of document navigation in Greenberg, S. Revisiting Read Wear: Analysis, Design, and Evaluation of Microsoft Word and Adobe Reader, and draw three main a Footprints Scrollbar. Department of Computer Science and Software conclusions: that region revisitation is frequent, that Engineering. Tech Report #02/08. 2008. current revisitation tools other than the normal scrollbar

are used very little, and that short recency lists can immediate suboptimal strategies are favoured over more contain most locations that people will revisit. efficient long-term ones) suggests that people will often fail 2. Can scrollbar marks improve revisitation, and how many to place bookmarks, even when they can foresee the long marks should be used? We conducted two experiments to term advantage [1]. This is why our research interest lies test the performance of adding marks into the scrollbar. primarily in automatic revisitation support. One study demonstrates that marks can decrease Automatic revisitation tools have been developed to support revisitation time; the other shows that people can revisitation both between and within documents. Familiar successfully use a large number of marks. interface controls such as history lists, ‘Back/Forward’ 3. How can a realistic interface support revisitation? We buttons, and ‘Recent Document’ menus facilitate navigation designed our Footprints Scrollbar (Figure 5) to maximize between documents. These automatic revisitation tools support for revisitation, based on principles derived from minimally intrude on users’ activities: they silently record the log analysis and the two studies. actions, populate a data structure or visualisation, and 4. Is this support used effectively and is it preferred? We provide support when called on. Their primary evaluated the Footprints Scrollbar in two experiments: a disadvantages are that people may not understand the controlled lab study and a realistic usage study. The algorithm for recording or presenting events, and the event Footprints Scrollbar was frequently and successfully set may overwhelm or fail to match the interests of the user. used, improved users’ revisitation time, and was strongly For example, people often misunderstand the behaviour of preferred to traditional scrollbars. the web ‘Back’ button, causing frustration when items Overall, our investigations show that revisitation is worth cannot be revisited [12], and we have observed similar supporting, and that read wear – long known but little problems with missing items in ‘Recent Documents’ studied – is a valuable and usable technique for improving menus, and with user misunderstanding of the temporal revisitation. To our knowledge these studies are the first to order of Alt-Tab window switching. analyse real world revisitation in documents and to use the There are also several widely-deployed examples of results in the design, implementation, and evaluation of a automatic revisitation tools for navigating within read-wear scrollbar. documents. For example, the web browser’s ‘Back’ button works as normal when navigating through internal page RELATED WORK links, and Adobe Reader’s ‘Previous/Next View’ feature Revisitation and its support steps through a linear history of scroll positions and zoom Zipf’s Law [33], the Pareto Principle [22] and the “80-20 levels. Visual Studio also has a ‘previous/next’ list of lines Rule” all demonstrate that many facets of human behaviour that the I-beam cursor has visited. These history lists leave are highly repetitive: although we have extensive no visible trace in the scrollbar, however, so people cannot vocabularies, wardrobes, or choices at our disposal, most of visually scan potential target regions without displaying the time we reuse a small set of favoured options. additional windows or menus. Researchers have also demonstrated that many aspects of computer use are also highly repetitive, such as command Hill et al ’s [18] original Read Wear system showed a use [15, 16], menu selections [11, 14], and web page visits histogram overview of the reading history of an entire [2, 12, 29]. document within the scrollbar. Each horizontal line of pixels in the scrollbar encoded information such as the These observations have stimulated research into interface number of edits, or length of time reading. Similar scrollbar techniques for supporting reuse and revisitation. Some marks are used by several code editors, but the marks are strategies require manual intervention while others used to highlight semantic information such as compilation automatically observe actions and update the interface. errors rather than to support revisitation. Manual revisitation tools require an explicit user action to The concept of document read wear inspired several mark information as interesting. For example, bookmark researchers to examine a variety of techniques for recording tools in web browsers and Microsoft Word let people and visualizing activity beyond the scrollbar. These include explicitly create iconic labels as shortcuts for returning to Wexelblat and Maes’ [30] ‘Footprints’ system, which particular pages or positions in a document. The Bookmark provided maps, trails, annotations and signposts for Scrollbar [24] is similar, but bookmarks are placed within a information foraging, and Skopik and Gutwin’s [27] ‘visit standard scrollbar. However, bookmarks have problems that wear’ marks for revisitation in fisheye visualizations. limit their use in practice [1]. First, they depend on people knowing in advance that the information will be required in To our knowledge, the idea of combining visitation the future. Second, people must believe that adding a histories with overview visualizations in the scrollbar has bookmark will yield benefits that exceed the manipulation not been pursued since Hill et al’s initial investigation. costs of creating and managing them. Third, people’s propensity for adopting satisficing strategies [26] (where Scrolling Enhancements ‘visit’, with little difference to the results. We report results Many researchers have investigated ways to improve only for two-second pauses. scrolling. Four general approaches have been: to improve the control device (e.g. isometric joystick [32] or scroll- When revisiting document locations, people are unlikely to wheel [19]); to improve the control-feedback loop (e.g. arrive at precisely the same viewport as the one previously speed-dependent automatic zooming [13, 21] or seen. We therefore define a revisit as returning to anywhere OrthoZoom [5], both of which use zooming to speed long within the bounds of an earlier visit and remaining there for distance movement); to improve spatial memory of more than two seconds. Furthermore, revisits are only document regions (e.g. Space-Filling Thumbnails [10] or logged when the person visits another position outside the Stationary Scrolling [28]); and to augment the scrollbar current view prior to returning. This condition ensures that with meaningful information (e.g. Edit/Read Wear [9, 18], actions such as slowly advancing line by line are not logged bookmarks [24], or the location of search terms [6-8]). as revisits.

Our research bridges the third and fourth approaches, using Patterns of Revisitation an augmented scrollbar to reinforce spatial memory and aid A large proportion of documents were opened and closed revisitation. First, however, we report the findings of our without any navigation: 39.6% and 19.6% for Word and log analysis of behaviour in document revisitation. Reader. This is probably due to people launching the application without an associated document (e.g., via the LOG ANALYSIS OF REVISITATION Start Menu), or due to opening the wrong document and How do people currently revisit document locations? immediately closing it. Removing these navigation-free Knowing actual revisitation patterns is critical for the instances from the analysis shows that 32% of Word and design of a revisitation system. To answer this question, we 27% of Reader documents include at least one revisit analysed logs of all navigational actions in Microsoft Word (row 9, Table 1). Row 10 shows the mean of participants’ 2003 and Adobe Reader 7 as used by 14 participants (all maximum daily revisits: 188.8 (Word) and 32.3 (Reader), computer scientists) during their normal work over 120 indicating that revisitation is a frequent activity in some days. The logging software and initial data analysis are documents. reported in [3, 4]; the analysis of revisitation discussed below is new. The logs precisely record the location and Use of existing revisitation tools extent of the scroll thumb every 200ms, describing the None of the participants used Word’s bookmarking user’s viewport independent of zoom level. Note that in function or Reader’s ‘Next View’ function during the 120 large documents (where the scroll thumb is the minimum days. A single participant used Reader’s ‘Previous View’ allowable size) the logged viewport exceeds the actual one. function in two separate documents (0.3% of his Table 1 shows summary statistics of the documents used documents). During interviews following the log analysis, during the study. all but two of the participants stated they were not aware of Microsoft Word Adobe Reader the Previous/Next view features. 1. Total documents 2342 1706 2. Number of documents 167±122, 37-384 122±183, 5-718 Word’s ‘Split window’ feature is another tool that can 3. Document page length 6.3±4.5, 1-160, 38±35.7, 1-1743 support revisitation; it allows users to simultaneously view 4. Number of days used 41.6±21.6,16-81 25.6±15.1, 4-53 more than one document region, eliminating the need for 5. Mean docs opened/day 3.7±1.6, 1.8-7.3 4.1±5.6, 1.3-23.1 6. No navigation in document 928 (39.6%) 334 (19.6%) repetitive scrolling between two regions. Seven of the 7. Document scrolled 1414 (60.4%) 1372 (80.4%) fourteen participants used split windows at least once, with 8. No revisits 969 (41.4%) 998 (58.4%) a total of 42 invocations. One participant accounted for 9. >1 revisit (% of scrolled docs) 445 (31.5%) 374 (27.3%) more than half (24) of these events. 10. Mean max. daily revisits 188.8±114.0 32.3±28.3 Table 1: Document usage and revisitation with Microsoft Recency and frequency list coverage by length Word and Adobe Reader. Rows 2-5 show mean, standard Hill et al .’s [18] read wear system displayed the amount of deviation (as ±), and the min-max range across users. time spent on every line, which is useful for returning to Defining Visits and Revisits frequently used locations, but not for recent locations. This When navigating within documents, people frequently stop raises the question of whether recency or frequency is a over a region without any intention to do so, caused by more effective basis for revisitation support. mechanical demands such as clutching the mouse or scrollwheel, or by cognitive and perceptual issues such as One key question for either design is that of how many determining whether the document is correctly positioned. items should be displayed in the scrollbar: e.g., the two We therefore need to define a specific pause length as most recently/frequently visited, or top twenty? Increasing denoting an intentional visit , and to prune visits of shorter the number of marked items increases the number of duration. The analyses that follow were all repeated with positions available for revisitation, but it also increases the one, two and three second stationary pauses used to define a user’s search load when scanning for a target. We therefore

analysed the logs (using methodology from previous studies on it (Figure 2). We then conducted an initial experiment to of revisitation [15,29]) to determine the proportion of determine whether people would use the recency marks (or revisited items covered by a recency or frequency list of ignore them), and whether using them would aid their length n. revisitation performance. Figure 1a shows the results for recency list analysis. For Tasks involved searching for locations in a plain text example, a one-item recency list (e.g., a simple ‘Back’ document (Joyce’s Ulysses , obtained from gutenberg.org) button) would allow users to reach 19% and 28% of and periodically revisiting them. The experiment also previous locations with Word and Reader. Longer lists investigated how performance was influenced by the rapidly increase the proportion of accessible locations, with number of marks in the scrollbar and by the number of ten-item recency lists covering 81% of revisitations with revisits to the same location (the latter providing insight Word and 84% with Reader. into how well users learn locations). Users completed the tasks using a simple document-viewing application that Frequency list analysis (Figure 1b) shows a similarly steep allowed navigation only with the scrollbar. Two versions slope, with 10 items allowing access to slightly fewer items were created, whose interfaces differed only in scrollbar than the recency list, at 78% and 83% of revisited locations type: a standard scrollbar, or a marking scrollbar that for Word and Reader respectively. The values for a one- showed red marks for visited locations. Middle-clicking on item frequency list are also lower than recency lists at 16% a mark immediately scrolled the view to its associated and 13% respectively, meaning that fewer regions would be location. In this experiment, marks were only placed when accessible with a simple ‘Back’ button than with recency targets were successfully visited or revisited; pausing lists. elsewhere in the document did not place a mark. Consequently, this is a best-case study for marking scrollbars; it considers questions about whether people will use the marks, whether the marks improve performance, whether the marks distract from other tasks, and how well users learn document locations both with and without marks.

(a) Recency list. (b) Frequency list. Figure 1: Percent of revisited items available through theoretical recency and frequency lists of different sizes.

Summary of Log Analysis The log analysis showed: that people frequently revisit Figure 2. Experiment one interface. Marking scrollbar (left) document locations; that current revisitation tools in Word and task cueing interface (right). Image altered for clarity. and Reader are seldom used; that relatively short revisitation lists provide good coverage of the locations that Participants and Apparatus users revisit; and that recency-based lists provide better Twelve volunteer university students (seven female) coverage than frequency based lists. This last point is participated in the experiment. Their mean age was 24 years intuitively reasonable – users’ interests in document regions (s.d. 5 years), and they were all experienced users of change over time; for instance, the morning’s work at window-based software (>10 hours/wk). The experiment regions x and y may be unimportant during the afternoon’s lasted approximately 30 minutes. work at regions w and z. A Java-based system was built for the study. A single window displayed both the experimental interface EXPERIMENT ONE: CAN RECENCY MARKS HELP? (1485×800 pixels, Figure 2 left side) and a task cueing pane The log analysis suggests that a recency-based revisitation (Figure 2, right). The task-cuing pane displayed an initial mechanism is worth pursuing. We designed a simple direction to the next target (up or down), the text of the marking scrollbar that displayed recency-based scroll marks sentence to be located, and a start/finish task button. The experiment ran on a standard Windows PC with a 22” LCD 11.0s) than traditional scrollbars (20.4s, s.d. 14.8s): monitor. F1,11 =33.8, p<.001. As anticipated, there were also significant main effects for revisit iteration (F =10.8, Experimental Design and Procedure 3,33 positions visited The task-time dependent measure is analysed using a p<.001), as indicated in Figure 3, and for 2×2×4 repeated measures analysis of variance (ANOVA), (F1,11 =21.2, p<.005), with mean acquisition times increasing with the three factors: interface type (standard and marking from 13.1s with five visited locations to 18.6s with ten. scrollbars), positions visited (5, 10) and revisit iteration (1st , There was a significant interface type × positions visited 2nd , 3 rd , or 4th revisit). interaction (F1,11 =5.3, p<.05), with the marking scrollbar showing greater benefits with more positions visited. There Participants were given a brief introduction to each was no interaction between interface type and revisit interface and the experimental method before completing iteration : F3,33 <1. three practice tasks with their first interface. They then 30 completed all tasks with one interface before proceeding to Marking scrollbar the next, with the order counterbalanced. Participants Standard scrollbar completed NASA Task Load Index (TLX) worksheets and 25 provided comments after completing each interface. 20 Tasks involved using the experimental interface to locate sentences displayed one at a time in the cueing interface. 15 Participants were informed that all target sentences were the 10 first in a paragraph and that tasks were completed by placing the target sentence anywhere within the viewport 5 and clicking the “Finish Task” button. Successfully Mean selectiontime (seconds) completing one task automatically cued the next. 0 1 2 3 4 Two sets of ten tasks were generated: the first set within the Revisit iteration first twenty pages of the document, and the second set Figure 3. Mean acquisition times in experiment one for the within the second twenty pages. All participants used the two scrollbar types across revisit iteration. first set with their first interface. The targets were generated to have similar locations across sets (within one paragraph Subjective measures and participant comments supported of each other). Targets were spaced nearly evenly through these positive results for marking scrollbars. NASA-TLX the 20 page regions, with at least one window of text worksheet results (ratings from 1=low to 5=high) showed between them. Equivalent targets in both sets were revisited lower mean workload and higher mean performance ratings the same number of times. Participants were unaware of for the marking scrollbar in all categories, although only the these constraints on target placement. overall ‘Effort’ measure showed a significant effect (see Table 2). The set of targets were visited as follows. First, participants found five consecutive targets for the first time, progressing NASA-TLX Standard Marking Sig? Mental load 2.9 (1.1) 2.4 (0.9) 0.08 downwards through the document: t1, t 2, t 3, t 4, t 5. Next, they Physical load 2.1 (1.4) 1.7 (0.7) 0.14 entered the first of two revisitation phases, revisiting items Temporal load 2.4 (1.0) 2.3 (1.1) 0.32 Performance 3.3 (1.2) 3.7 (0.9) 0.25 in the order t2, t 4, t 2, t 1, t 2, t4, t 2. They then visited the five remaining targets for the first time ( t –t ), resulting in ten Frustration 2.5 (1.2) 2.2 (1.1) 0.31 6 10 Effort 2.9 (0.9) 2.0 (0.9) 0.02 marks in the scrollbar, followed by a second revisitation phrase with target order t3, t 9, t 3, t 6, t 3, t9, t 3. In total, each Table 2: Mean (st. dev.) NASA-TLX responses for standard participant completed 48 experimental tasks: 2 targets and marking scrollbars in experiments one. Wilcoxon Signed Ranks significance test. visited 5 times each (t2 and t3), two visited three times ( t4 and t9), two visited twice ( t1 and t6), and four visited once Participant comments were positive: “The scrollbar mark is (t5, t7, t8 and t10 ), giving 24 tasks. These were then repeated extremely nice”. One participant noted that the marks for the second interface. helped reduce their spatial search space: “I would usually Tasks were automatically completed after 90 seconds to only be off by one mark if I didn't choose correctly the first reduce the impact of situations where participants became time, so it was easy to correct my mistakes.” lost. Six tasks were discarded due to exceeding the time Discussion limit: two with marks, and four with traditional scrollbars. These results show that accurately placed scrollbar marks helped participants return to document locations, and that Results the support was appreciated. Also, as more places were Figure 3 summarises these results. The marking scrollbar visited, participants gained more from the marks. allowed significantly faster revisitation (mean 11.3s, s.d.

There are three questions that are yet to be addressed in the is provided by regression analysis of acquisition time with investigation: first, how is performance influenced by large positions visited , which gives a logarithmic best fit with numbers of visited locations (and therefore marks); second, R2=0.98. The Hick-Hyman Law [17, 20] shows that what is the performance of the marking scrollbar when logarithmic functions relate decision time to number of marks are automatically placed wherever the user pauses, items, while linear functions apply to serial candidate rather than the idealized placement on targets tested here; searches. and third, how does the scrollbar work with naturalistic 25 revisitation rather than artificially-organized targets? 5 10 20 30 Experiment two addresses the first of these issues: how is 20 performance influenced by large numbers of visited locations. 15

10 EXPERIMENT TWO: MARKING OVERLOAD

The log data showed that 10 marks covers 80% of 5 revisitations, and that 30 marks covers close to 100%. Yet as Figure 1 reveals, each additional mark covers a smaller Mean acquisition time (seconds) 0 proportion of revisitation targets, therefore offering 1 2 3 4 progressively lower utility while increasing the number of Revisit iteration Figure 4. Mean revisit times with 5, 10, 20 and 30 scrollbar distracters. This study, therefore, investigates how st th performance is influenced by the number of marks. marks on the 1 to 4 revisit in experiment two. These results suggest a tradeoff between the number of Experimental Method marks and performance. Although the participants quickly Apparatus and participants . The experiment used the same learned many marks and their associated regions marking interface, cuing interface, and apparatus as (performing well by the 4 th revisit), acquisition times on experiment one. In addition, the same people from the first their first revisit increased steeply with the number of experiment participated in the second, and advanced to the marks (Figure 4, far left). Nevertheless, these results show second experiment after a short break. that large mark sets are feasible, and that designing to cover nearly all revisits is a possibility. Since ten marks cover Design. Revisitation time is analysed using a 4×4 repeated more than 80% of revisited locations, however, (see Figure measures ANOVA with two factors: positions visited (5, 1), we use 10 marks in our remaining studies. 10, 20, 30) and revisit iteration (1 st , 2 nd , 3 rd , 4th ). THE FOOTPRINTS SCROLLBAR Tasks. Participants completed three preparation tasks and The results of the first two studies, and our experiences then searched and revisited locations within the first 45 with the initial version of the marking scrollbar, led us to pages of James Joyce’s Dubliners using a similar procedure design a new version – called the Footprints Scrollbar to that used in experiment one (marks were automatically (Figure 5) – that contains a number of more advanced placed on targets when acquired). features for supporting revisitation.

Results and Discussion The Footprints scrollbar supports six related methods for Figure 4 summarises the results. As anticipated, there are revisiting regions. First, coloured marks are placed in the significant main effects for both number of positions visited scrollbar to provide spatial cues to previously visited areas. (F3,33 =13.4, p<0.001) and revisit iteration (F 3,33 =22.4, Marks gradually fade from ‘hot’ colours (reds, oranges) p<0.001). There is also a significant interaction between the through to ‘cold’ one (greens, blues) to denote their factors (F 9,99 =1.99, p<0.05), which is best explained by the increasing age in the recency set. Second, middle clicking large absolute task time reduction between iterations 1 and on a mark causes a rapid, animated scroll transition to the 4 with 30 marks compared to the smaller reductions with associated view [23]. Third, when the user moves the cursor fewer marks. This should be expected, since learning and over the scrollbar, small thumbnail images quickly fade into remembering thirty marks is clearly more demanding than view alongside each mark, giving a visual overview of the five. associated document regions. Moving the cursor over a By the fourth iteration, performance appears to be reaching thumbnail expands that thumbnail for better visual an asymptote for all conditions, with little performance inspection (an example is shown at the bottom of Figure 5). difference between 10, 20 and 30 marks (Figure 4, far The thumbnails fade out when the cursor moves away from right). This suggests that participants were not overloaded the scrollbar. Clicking a thumbnail also moves to that view. by 30 marks. Fourth, back/forward functions are invoked by the ‘left’ and ‘right’ keyboard keys; this allows users to rapidly move Additional evidence suggesting that users were successfully through the mark history and its corresponding region “deciding” about marks rather than linearly searching them views. Fifth, depressing the Shift key and rotating the number, or dragging the thumb to the mark are all equivalent methods of revisitation. The two-second timeout has an important impact on the behaviour of the interface: it determines when marks are placed, and determines the semantics of the back/forward keys. To help users predict and comprehend this behaviour, the scroll-thumb depicts the approaching timeout by gradually filling with colour, similar to a progress bar (see right). Once filled, the timeout expires, a mark is left in the scrollbar, the region is inserted at the tail of the recency list, and the window view is captured for use in the associated thumbnail. To precisely describe this behaviour, Table 3 uses comma separated letters a, b, c… to represent visited document regions; the symbol ⊥ denotes a 2 second static location; subscripts 1, 2, 3… denote shortcut digits on marks; and denote the back and forward keys; and curly braces {…} Figure 5: The Footprints scrollbar. enclose the comma-separated content of the recency list, scrollwheel moves the document position to the closest with the most recent item at the right hand end of the list. mark in the direction of rotation (according to distance, not An underlined item in the recency list represents the region recency). Finally, marks are overlaid with the numbers 1– displayed at the end of each action sequence. 10: users can trigger rapid movement to the selected Actions Region Recency list position through the numeric keypad or by typing a number sequence and marks into a ‘go-to’ box. Number assignment is arbitrary, with the 1. Visit three regions a⊥, b ⊥, c ⊥ {a 1, b 2, c3 } ⊥ shortcut remaining constant for the life of the mark. Details 2. Scan 3 regions then pause d, e, f {a 1, b 2, c3, f4} 3. ‘Back’ three times (no pauses) , , {a1, b 2, c3, f 4} of these features follow. 4. ‘Forward twice (still no pauses) , {a 1, b 2, c3, f 4} ⊥ 5. Pause {a 1, b2, f4, c3} 6. ‘Back’ and pause ⊥ {a , b , c , f } Marking Algorithms and Behaviour 1 2 3 4 7. ‘Back’ and pause ⊥ { a 1, b2, f4, c3} Each mark is 16 pixels high (equal to the default minimum Table 3: Navigational behaviour of the Footprints Scrollbar. size for a scroll thumb in most widget sets), and marks never overlap one another. Marks are only placed in the The timeout allows users to dynamically traverse the scrollbar when the scrolled region remains static for more recency list with the ‘Back’ and ‘Forward’ shortcut keys: than two seconds. Consequently, both continuous scrolling for example, quickly flipping between regions (e.g. rows 3 and scrolling with short pauses for device clutching or and 4 of Table 3). However, when editing two document visual inspection have no impact on the marks and the regions (e.g. ‘Results’ and ‘Abstract’) users are likely to recency list. spend relatively long periods in both areas, and they can move between them with subsequent presses of ‘Back’ (e.g. Scrollbar marks are produced from a recency list data rows 6 and 7). Experiment four investigates the usability of structure that removes duplicates [15, 29]. Whenever the this model, and how users chose to interact with the system; scroll view remains static for more than two seconds, that first, we report on an experiment that tested the region is inserted at the end of the recency list, and any performance of the new system. earlier entry for the same location is removed from the list. The current implementation has a 1:1 correspondence EXPERIMENT THREE: FOOTPRINTS EVALUATION between marks and list entries, but this could be relaxed to Log analysis showed that people could benefit from allow more items on the recency list than are visualized in revisitation support. The experimental results above suggest the scrollbar: e.g., providing 10 visible marks, but allowing that people can benefit when using even a simple marking a larger number of previously visited regions, accessible via scrollbar, if marks are correctly placed. Correct mark the Back/Forward keys. placement is an artificial ideal, however, and realistic mark Marks are numbered from 1 to n, where n is the placement needs to be evaluated. configurable maximum (with a default of 10). To reduce This experiment compared the Footprints scrollbar with a visual distraction associated with scrollbar changes, the standard scrollbar in terms of people’s performance and numbers are as stable as possible, with each mark after n preferences. Tasks involved finding and revisiting receiving the mark number of the least-recent member of document regions, with the Footprints system automatically the recency list. Clicking the mark, typing its shortcut placing marks whenever the user paused for more than two

seconds. Unlike experiment one (which controlled the shows the type×length interaction – both interfaces number of revisits to each position), this experiment performed similarly with 10 page documents (means of 6.9s controlled revisits according to mark position within the vs. 6.6s), but Footprints performed better with 40 page recency list. This allowed us to inspect how performance of documents (mean 7.9s vs. 10.9s). The type×position the Footprints scrollbar was affected by the differing ages interaction is due to the Footprints scrollbar out-performing of different marks. The disadvantage of this method is that standard scrollbars in all but the 11 th position on the recency it artificially made less-recent revisits more likely than our list. On trials in the 11 th list position, the Footprints system logs show them to be. provided a corresponding mark only 15% of the time (due to marks falling off the end of the recency list), so users Experimental Method who tried to use marks would have been misled. Means on Twelve participants (two female) took part in the the 11 th trial were 12.0s and 10.5s for Footprints and experiment. Tasks involved finding and re-finding standard scrollbars. document regions that were cued by displaying an image of the target region and an initial direction (up or down). 12 Footprints scrollbar Participants began tasks by pressing a “Start Task” button, 10 Standard scrollbar and completed them by scrolling the target region into the 8 middle two-thirds of the screen and clicking a “Finish Task” button. A red status message was displayed if the 6 target was not correctly positioned, and the task continued. 4 Training with each interface was similar to experiment one, 2 and was followed with eight sample tasks to familiarize

Mean acquisition time (secs) time Meanacquisition 0 participants with the procedure. 10 40 Document length All participants used the Footprints scrollbar and the standard scrollbar (in counterbalanced order) with ten- and Figure 6: Experiment three results, by document length. forty-page documents (two papers from the CHI proceedings and two instruction manuals). Eleven target All but one participant preferred the Footprints scrollbar; regions were generated for each document by evenly the remaining participant could not choose between them. distributing preliminary locations, then randomly adjusting Mean NASA-TLX worksheet results, were uniformly better these locations by between 0 and 5%. for Footprints: significantly so for ‘Physical Load’, ‘Performance’ and ‘Effort’. Participant comments from The eleven targets were initially presented consecutively experiment three are discussed below, together with those from the top of the document to the bottom. Participants of experiment four. then revisited targets according to their ideal position on the recency list: three times for each position 2–11. The ideal EXPERIMENT FOUR: OBSERVATIONAL STUDY positions assume that marks only fall on targets. However, Tasks in experiment three were tightly constrained for marks were placed whenever the user paused, so the ideal experimental control, but they artificially induced recency list is likely to differ from the user’s actual one. revisitation and artificially exaggerated temporally distant Therefore, latter list positions may not have been visible in revisitations. To inspect more realistic tasks as well, our the scrollbar when needed, due to being displaced by other final study used a structured interview process to observe marks. Analysis of the study logs showed that the location participants’ more natural interaction within a document of in 8 th position on the ideal list was in the user’s visible list their choosing. 90% of the time, the 9 th 72%, the 10 th 52%, and the 11 th Eight participants (one female), all graduate students or 15%. faculty from Computer Science, took part in the experiment. They were asked to identify a favourite Results Task time data are analysed using a 2×2×10 RM-ANOVA research paper, which we loaded into the system. for factors interface (Footprint, standard), document length The experiment began with a two-minute introduction to (10 and 40 pages) and recency list position (2–11). the system’s features, explaining marks, thumbnails, digit There was a significant main effect for interface type shortcut keys, and the back/forward arrow keys. Participants then completed 25 tasks using whichever (F 1,11 =8.24, p<0.05), with Footprints (7.4s) faster than standard scrollbars (8.8s). Document length (F =69.6, methods they chose. Tasks involved describing the paper to 1,11 the experimenter in response to a consistent set of p<0.001) and recency list position (F 1,11 =18.8, p<0.001) both showed expected main effects. Importantly, there were questions: “show me what you think is the best part of the paper”, “who is on the reference list?”, “where is related significant type×length (F 1,11 =15.1, p<0.01) and type×position (F =2.9, p<0.01) interactions. Figure 6 work summarised?”, “what’s the first paper referenced in 10,110 the Introduction”, “where was that paper published”, etc. Participants then completed a questionnaire on the denoting mark age) with smaller marks and higher levels of effectiveness of various system components. Automatic opacity to ensure they do not interfere with thumb logs captured all user actions with the interface. acquisition. Digit shortcuts. Only one of the participants used digit Results Questionnaire responses (summarized in Table 4) show that shortcut keys for navigation, and another one commented participants found marks and thumbnails very helpful for that he used the digit marks ‘to map locations’. Generally, visualizing and navigating to previously visited locations, though, the digit marks were criticized by a few participants with Likert scale (1=disagree; 5=agree) means ranging as either unnecessary or mildly confusing. between 4.5 and 4.9 (rows 1–3). These positive ratings were Back/forward key shortcuts. Most participants stated that achieved without substantial distraction (mean distraction they understood the behaviour of this mechanism (mean ratings of 2.3 for marks and 1.5 for shortcut numbers). 4.5) and that they might use the keys in other documents, Overall, comments were highly positive, including but none of the participants actually used them during the “revisitation is tremendously useful and would probably tasks. One mentioned a conceptual clash between the only improve as the document increases in size”, “really ‘Forward’ key and ‘forward=down’ in the document, useful”, and “it’s additive: no interference with any other despite understanding the behaviour. Another stated “using widget”. Five of the participants stated that they would back/forward arrows is something that just didn’t occur to want all of the supported features in their desktop me.” This comment echoes the findings of the log analysis interfaces; three stated they would want some of the – current recency tools, such as Reader’s ‘previous/next features, with all wanting the thumbnails, region markers, view’ or Word’s bookmarks are not sufficiently ready-to- and middle-click shortcuts. hand, and hence they go unused despite their potential Question Mean (s.d.) utility. 1. Marks were helpful to see where I’d been. 4.8 (0.5) Lack of control over mark placement. Some participants felt 2. Marks were useful for revisiting locations 4.9 (0.4) 3. Thumbnails were useful for revisiting locations 4.5 (1.1) pressured by the thumb-filling animation, with one 4. I understood mark placement 4.9 (0.4) commenting that it “made me rush before it dropped a 5. Marks were distracting 2.3 (1.2) mark” and another stating that “it would be nice to 6. Colours were useful 2.1 (1.0) somehow stop the dropping of the marks”. The 7. Back/Forward keys were useful 2.8 (1.3) 8. I understood the Back/Forward keys 4.5 (1.1) questionnaires asked participants to comment on whether 9. Shortcut numbers were useful 2.5 (1.3) the two second marking timeout was too short or too long, 10. Shortcut numbers were distracting 1.5 (0.8) with several responding that it was “sometimes too short Table 4: Mean responses to five point Likert scale questions in and sometimes too long”. Lightweight controls for experiment four. manually adding and removing marks in the scrollbar could Discussion of Experiments Three and Four solve this problem, but it is unclear whether people would Experiments three and four show the value of the Footprints use such controls. scrollbar for revisitation in both a controlled, and a more naturalistic situation, especially for longer documents. This CONCLUSIONS success validates our empirical recommendation that Revisitation has been comprehensively investigated in revisitation should be supported (log activity analysis), and domains such as web navigation and command use. our initial evaluations of scroll marks in artificial situations. Somewhat surprisingly, however, region revisitation within People also preferred the features of the Footprints documents has been largely overlooked. Our log analysis scrollbar. demonstrates that users frequently revisit document regions and that short revisitation lists can theoretically provide Despite these successes, the Footprints scrollbar is still an access to most locations that users return to. We used these early design. Although the overall system was praised, findings to motivate and inform the design of a system, participants identified several areas for improvement, all of based on Hill et al.’s read wear, that augments the scrollbar which can be incorporated in the next design iteration. with marks that aid revisitation; a series of evaluations Mark colours . Nearly all participants commented that demonstrate that the system can improve user performance. colour poorly communicated mark recency. Worse, three Our Footprints scrollbar works within the current ecology participants observed that their memory for items was of graphical user interfaces – it augments the familiar harmed by colour changes – they might remember the ‘red’ scrollbar rather than replacing it, and it occupies the same mark in a region, only to be confused by later colour location on the screen. Except for shortcut keys (which can reconfiguration. Furthermore, two participants noted that be easily modified) the input to this scrollbar does not coloured marks increased the difficulty of visually compete with input actions that control other parts of a acquiring the scroll thumb. We therefore recommend that document viewer or GUI. While improvements were future implementations use stable mark colours (not

suggested through experimental feedback, these can be 15. Greenberg, S. and Witten, I. Supporting Command easily incorporated in future system refinements. Reuse: Empirical Foundations and Principles. Int. Journal of Man-Machine Studies 39 (1993), 353--390. Our future work with the Footprints scrollbar will progress 16. Greenberg, S. and Witten, I. Supporting Command in two main directions. First, we plan to revise the design of Reuse: Mechanisms for Reuse. Int. Journal of Man- the scrollbar according to the user responses discussed Machine Studies 39 (1993), 391--425. above, and produce a version that can be incorporated into 17. Hick, W. On the rate of gain of information. Quarterly real-world document readers. Second, we will carry out J. Experimental Psychology 4 (1952), 11-36. longer-term evaluations of the device, to provide additional 18. Hill, W., Hollan, J., Wroblewski, D. & McCandless, T. information about how revisitation and revisitation support Edit Wear and Read Wear. Proc. CHI'92. (1992) 3-9. work in real use. 19. Hinckley, K., Cutrell, E., Bathiche, S. and Muss, T. Quantitative Analysis of Scrolling Techniques. in Proc. REFERENCES CHI'02. (2002), 65-72. 1. Abrams, D., Baecker, R. and Chignell, M. Information 20. Hyman, R. Stimulus information as a determinant of Archiving with Bookmarks: Personal Web Space reaction time. J. of Exp. Psych. 45 (1953), 188-196. Construction and Organization. in Proc. CHI'98 , (1998), 21. Igarashi, T. and Hinckley, K. Speed-dependent 41-48. Automatic Zooming for Browsing Large Documents. in 2. Adar, E., Teevan, J. and Dumais, S. Large Scale Proc. UIST'00. (2000), 139-148. Analysis of Web Revisitation Patterns. in Proc. CHI'08. 22. Juran, J. Quality Control Handbook . McGraw-Hill, New (2008), 1197-1206. York, 1951. 3. Alexander, J. and Cockburn, A. An empirical 23. Klein, C. and Bederson, B. Benefits of Animated characterisation of electronic document navigation. in Scrolling. Ext. Abstracts CHI'05. (2005), 1965-1968. Proc. Graphics Interface (2008), 123-130. 24. Laakso, S.A., Laakso, K.-P. and Saura, A.J. Improved 4. Alexander, J., Cockburn, A. and Lobb, R. AppMonitor: Scroll Bars. Ext. Abstracts CHI'2000 , (2000), 97–98. A Tool for Recording User Actions in Unmodified 25. O'Hara, K. and Sellen, A. A comparison of reading Windows Applications. Behavior Research Methods 40 , paper and on-line documents. in Proc. CHI'97. (1997), 2 (2008), 413-421. 335-342. 5. Appert, C. and Fekete, J. OrthoZoom Scroller: 1D 26. Simon, H. Satisficing. in Eatwell, J., Millgate, M. and Mutli-scale navigation. Proc. CHI'06. (2006), 21-30. Newman, P. eds. The New Palgrave: A Dictionary of 6. Beaudouin-Lafon, M. Instrumental Interaction: An Economics , Stockton Press, New York, 1987, 243-245. Interaction Model for Designing Post-WIMP User 27. Skopik, A. and Gutwin, C. Improving Revisitation in Interfaces. in Proc. CHI'00. (2000), 446--453. Fisheye Views with Visit Wear. in Proc. CHI'05. 7. Björk, S. The ScrollSearcher Technique: Using (2005), 771-780. Scrollbars to Explore Search Results. in INTERACT'01 , 28. Smith, R. and Taivalsaari, A. Generalized and stationary (2001), 745-746. scrolling. in Proc. UIST'99 . (1999), 1-9. 8. Byrd, D. A Scrollbar-based Visualization for Document 29. Tauscher, L. and Greenberg, S. How People Revisit Navigation. in Proc. Digital Libraries , (1999), 122-129. Web Pages: Empirical Findings and Implications for the 9. Chimera, R. ValueBars: An Information Visualization Design of History Systems. Int. Journal of Human and Navigation Tool. Proc. CHI'92. (1992), 293-294. Computer Studies, 47 , 1 (1997), 97-138. 10. Cockburn, A., Gutwin, C. and Alexander, J. Faster 30. Wexelblat, A. and Maes, P. Footprints: History-Rich Document Navigation with Space-Filling Thumbnails. Tools for Information Foraging. in Proc. CHI'99. in Proc. CHI'06. (2006), 1-10. (1999), 270--277. 11. Cockburn, A., Gutwin, C. and Greenberg, S. A 31. Wroblewski, D., Hill, W. and McCandless, T. Attribute- Predictive Model of Menu Performance. in Proc. Mapped Scroll Bars, US, 1990. CHI'07. (2007), 627-636. 32. Zhai, S., Smith, B. and Selker, T. Improving Browsing 12. Cockburn, A. and McKenzie, B. What Do Web Users Performance: A Study of Four Input Devices for Do? An Empirical Analysis of Web Use. Int. Journal of Scrolling and Pointing Tasks. in Proc. INTERACT'97. Human-Computer Studies 54 , 6 (2001), 903--922. (1997), 286-292. 13. Cockburn, A., Savage, J. and Wallace, A. Tuning and 33. Zipf, G. Human Behavior and the Principle of Least Testing Scrolling Interfaces that Automatically Zoom. in Effort: An Introduction to Human Ecology . Addison- Proc. CHI'05. (2005), 71-80. Wesley, Reading, Mass., 1949. 14. Findlater, L. and McGrenere, J. A comparison of static,

adaptive, and adaptable menus. in Proc. CHI'04 (2004), 89-96.